MEASURING AND FORECASTING CHINA`S AGRICULTURAL PRODUCTION, STOCKS, AND IMPORTS WITH HIGH QUALITY DATA AND MACHINE LEARNING METHODS

• Sponsoring Institution: National Institute of Food and Agriculture
• Project Status: ACTIVE
• Funding Source: AFRI COMPETITIVE GRANT
• Reporting Frequency: Annual
• Accession No.: 1031362
• Grant No.: 2022-67023-41013
• Cumulative Award Amt.: $649,980.00
• Proposal No.: 2023-06488
• Project Start Date: Sep 1, 2023
• Project End Date: Aug 31, 2026
• Grant Year: 2023
• Program Code: [A1641]- Agriculture Economics and Rural Communities: Markets and Trade

Recipient Organization
CORNELL UNIVERSITY
(N/A)
ITHACA,NY 14853

Performing Department
(N/A)

Non Technical Summary
In 2020, China had accounted for 20% of the United States' total agricultural exports. China'shuge and unexpected imports of US corn, soybean, and pork products may or may not continuein the next decade. There is a lack of reliable statistics on China's key agricultural commodities.This creates price volatility and might reduce the transparency of US and world agriculturalmarkets.We propose to develop reliable data/statistics for corn, soybeans, and pork. First, we will useprice, trade, and satellite data to get reliable statistics on production, consumption and stocks.Second, we will work with our partners in the Huazhong Agricultural University to conductplanting intention surveys in key commodity growing areas to understand the economic andpolicy determinants of Chinese crop producers' decisions.We propose to develop machine learning (ML) models to forecast China's agricultural imports.We will compare the prediction accuracies of ML models with traditional gravity models, whichleads to an understanding of the applicability of ML techniques in forecasting agricultural tradeand quantifying the impacts of trade and economic policies.We have shown the applicability of ML methods in predicting agricultural trade flows and alsoin predicting stock levels when some economic data is accurate, and some is not. The proposeddata collection and modeling system will provide insights into world agricultural trade patternsand enhance our ability to quantify future shocks to the global agricultural markets. The ultimategoal is to improve the long-term sustainability and resiliency of US agriculture and food systems.

Animal Health Component

75%

Research Effort Categories

Basic

25%

Applied

75%

Developmental

(N/A)

Classification

Knowledge Area (KA)	Subject of Investigation (SOI)	Field of Science (FOS)	Percent
606	6120	3010	75%
601	7299	2020	25%

Knowledge Area
601 - Economics of Agricultural Production and Farm Management; 606 - International Trade and Development;

Subject Of Investigation
6120 - International economy and sectors thereof; 7299 - Research equipment and methods, general/other;

Field Of Science
2020 - Engineering; 3010 - Economics;

Keywords

chinese agriculture

agricultural trade

crop production

machine learning

Goals / Objectives
The goal of this project is to gather and develop better quality, truth-grounded statistics on keyChinese agricultural commodities, coupled with suitable ML models, to better quantify andforecast Chinese agricultural production and imports with a focus on the impacts on United States (US) agricultural trade. We propose to address the three following objectives to be discussed in detailin the following sections:Objective 1:Develop reliable and granular statistics on production, consumption, and stocks ofmajor agricultural commodities in China by leveraging planting intentions surveys, satellite data,and ML methods (Lead: Zhang; Co-lead: He, Hayes, Xiong, Hu).Objective 2:Develop ML models, including supervised and unsupervised models, to fit andforecast China's overall agricultural production, consumption, and imports of major commoditiesover time (Lead: Hu; Co-lead: Hayes, He, Zhang).Objective 3:Disseminate better data on China's key agricultural commodities and validated MLmodels via an open-source platform that readily allows collaboration, contribution, andutilization by other researchers (Lead: He; Co-lead: Hayes, Hu, Zhang, Xiong).

Project Methods
Objective 1:Develop reliable and granular statistics on production, consumption,and stocks of major agricultural commodities in China leveraging plantingintentions surveys, satellite data, and machine learning methods.Task 1.1 Develop reliable statistics on corn and soybean production in China leveragingofficial and satellite data.First, we plan to collect and compare corn production and stock datafrom CNBSand USDA to identify key policy changes and events that make China revise its China's cornproduction and stock estimates (such as the Third Agricultural Census) and evaluate thejustification for these data revisions.Second, we plan to use satellite data to identify corn/soybean harvested area/production overtime. We will overlay the NPP high-resolution map at the 1km spatial level from NASA MODISproduct MOD17A3 with the cropland area maps to get corn/soybean production at the prefecturelevel over time.We will alsosubscribe to PlanetLabs' 3x3 meter PlanetScope data, which scans Earth's landmass every singleday and allows us to figure out the amount and location of cropland area.Third, we intend to compare corn and soybean production and area estimates from satellite datawith public prefecture-level data to identify administrative areas with potential data problems,and then evaluate these data inconsistencies to better understand the political and economiccauses of these inconsistencies, which could generate insights and better guide us in evaluatingthe authenticity of China's official agricultural statistics.Task 1.2 Generate reliable data on China's hog, soybean, and corn inventory.We first plan to collect comprehensive hog, soybean, and corn inventory, use, and price data atdifferent administrative levels from: (1) macro sources, such as CNBS and MARA; and, (2)micro sources, such as the Third Agricultural Census in 2018 and China's annual National RuralFixed-Point Survey data. We then aim to identify critical changes in the reporting baselines andpotential bias associated with these critical reporting changes. Finally, we aim to use MLtechniques to uncover more reliable hog, soybean, and corn inventory data based on price andcalibrated historic price-inventory and price-consumption relationships following Shao et al.(2021).Task 1.3 Collaborate with our partners in the Huazhong Agricultural University led by co-PIXiong to conduct planting intentions and planting surveys in key corn and soybean growingareas.Objective 2:Develop machine learning models, including supervised andunsupervised models, to fit and forecast China's overall agricultural production,consumption, and imports of major commodities over timeTask 2.1 Prepare the trade, policy, economic, geographical, and weather data used in machinelearning models.Task 2.2 Develop machine learning methods to identify and rank features affecting China'sagricultural imports.ML models will be developed, ranging from simple (linear regression) to more complex(ensemble) models to train on the data set with the objective of predicting China's import valuein the year 2019. The model development includes selecting the correct benchmarks based on theresearch objectives, developing an appropriate model based on the data requirements, trainingthe model on the training split while optimizing its usage of the validation split, evaluating themodel's predictive performance on the training and test splits, interpreting training and test errorsand possible overfitting issue, and comparing test prediction errors with the benchmarks.Finally we will interpret the model results, calculate and interpret the variable importance fromthe developed ML model, calculate and interpret the partial dependency plots of the developedML model, interpret the behavior of the model in predicting different commodities, interpret theability of the model in detecting unexpected trends, discuss the model complexity and its abilityto deploy to production, and evaluate the generalizability of the model by testing on some newunseen observations.Task 2.3 Use developed machine learning models to predict US-China agricultural trade.Objective 3:Disseminate better data on China's key agricultural commodities andvalidated machine learning models via an open-source platform that readily allowscollaboration, contribution, and utilization by other researchers.Task 3.1 Develop a web-based platform to disseminate estimates of China's corn and soybeanproduction, stock, and hog inventory at different administrative levels.We intend to build a web-based open-sourcedplatform to disseminate the data and models, and to allow users to freely download the data andcode to extend the model to other commodities.Task 3.2 Disseminate the ML methods and codes via a public platform to facilitate theapplication of ML methods in predicting agricultural trade flows in other major agriculturaltrading countries.Task 3.3 Organize a special session at the annual AAEA conference with outside researchersdiscussing these high-quality Chinese agricultural data and the potential of utilizing MLmethods to improve US agricultural trade projections.

Progress 09/01/23 to 08/31/24

Outputs
Target Audience:Agricultural producers, landowners, trade professionals and policymakers in United States (US) and China. Changes/Problems:Given the fraught US-China political relations, it is very difficult for foreign researchers to conduct planting intention surveys of Chinese farmers directly as we planned. As a result, we are proposing to mainly rely on satellite images and news report summaries of provincial-level planting intention surveys for objective 1. What opportunities for training and professional development has the project provided?A Visiting PhD student Miao Li from Huazhong Agricultural University helped PI Zhang in identifying data sources in China as well as applying machine learning methods in Chinese agricultural futures market analysis. PI Zhang has also identified a postdoctoral research associate (Jianwei Ai) Co-PI Hu hired two graduate RAs (Omid Rostami, Mohammad Fili) and one undergraduate RA (Jonas Swope) to help with objective 2. Co-PI Hayes hired one graduate RA (Xiaolan Wan) to assist with objective 1. How have the results been disseminated to communities of interest? Policy Briefs: Wang, F. and W. Zhang. 2024. Foreign Ownership and Leasehold of Agricultural Land in New York. Cornell University Dyson School of Applied Economics and Management Extension Bulletin EB 02-2024. Presentations Omid Rostami, Mohammad Fili, Jonas Swope, Guiping Hu, Wendong Zhang, Wei Zhang. "Industry Superstars: Unmasking Key Features that Drive Firm-Level Performance in Chinese Markets using Ensemble Learning with Genetic Algorithm". IISE Annual Conference, 2024. Zhang, Wendong. China's Agri-Food Trends. Invited Presentation for Iowa Corn Growers Association. Via zoom, March 5, 2024. Zhang, Wendong. China's Agri-Food Trends. Invited Presentation for Syngenta ALA Program, Center for Food and Agricultural Business, Purdue University. West Lafayette, IN, November 29, 2023. Zhang, Wendong. Foreign Interest in U.S. Agricultural Land. Invited Seminar for ASFMRA (American Society of Farm Managers and Rural Appraisers) National Conference, Nashville, TN, November 16, 2023 (Joint with Mykel Taylor) What do you plan to do during the next reporting period to accomplish the goals? Given the fraught US-China political relations, it is very difficult for foreign researchers to conduct planting intention surveys of Chinese farmers directly as we planned. As a result, we are proposing to mainly rely on satellite images and news report summaries of provincial-level planting intention surveys for objective 1. PI Zhang will also hire a postdoctoral research associate (Jianwei Ai) and a visiting PhD researcher (Ziyang Long) from Renmin University of China to help with objectives 1 and 2.

Impacts
What was accomplished under these goals? Objective 1: Develop reliable and granular statistics on production, consumption, and stocks of major agricultural commodities in China by leveraging planting intentions surveys, satellite data, and ML methods Co-PI Dermot Hayes and graduate RA Xiaolan Wan has a working paper "Forecasting Corn Acreage in China: Leveraging Political Incentives in Cropland Allocation": Using prefecture-level data from 2010-2021 in Northeast and North China, this paper explores the effect of government intervention on planted acreage of corn and estimates a corn acreage response function in China. We estimate an acreage response function where the share of corn area planted depends not only on expected profits but also on governmental factors. The regression results demonstrate the governmental role in crop acreage allocation. With a 1-percent increase in soybean subsidies, the corn acreage planted will decrease by 0.008 percent, holding all other factors constant. Moreover, government pressure for soybean expansion has a statistically significant negative effect on corn acreage. The results are compared with projections made by USDA. The results show that the updated models provide more accurate forecasts, with the mean square error reduced by more than half compared to the USDA forecast. The model is then used to predict 2025 corn planting and predicts a slight decline in corn acreage. PI Zhang, co-PIs He and Li have a publication at Journal of the AAEA that examines how the political alignments of Midwestern farmers, proxied by their consumption of partisan media, affect their perceptions of and responses to the US-China trade war. Our results indicate that farmers who consume conservative media perceive a lower income loss resulting from the trade war and view the Market Facilitation Program (MFP) as more helpful. Conversely, farmers who consume liberal media have the opposite perception biases. We found no evidence of any association between partisan media consumption and planting and risk management decisions. Overall, partisan bias exists despite financial interest at stake but does not affect behaviors. PI Zhang and co-PI Xiong has a publication that leverages data on daily stock returns from 20 listed Australian and 32 listed Chinese firms that produce the restricted commodities, we provide the first systematic analysis of the firm-level economic impacts of China's trade restrictions on Australian and Chinese firms. We find significant adverse effects on Australian firms' stock returns, leading to almost 20% loss within 10 trading days; however, most firms' stock returns immediately rebounded. In contrast, Chinese firms usually saw significant positive stock returns, leading to almost 30% gains, and the positive abnormal returns continuously increased within 10 trading days. Media coverage and trade dependence substantially impact Australian and Chinese firms' stock returns--industries with stronger trade dependence on China saw greater losses in Australian firms' stock returns. Co-PI He has one publication at China Economic Review "Dams, cropland productivity, and economic development in China": We use satellite, hydrology, and census data from 1992 to 2014 to quantify dams' impacts on cropland productivity and economic development in China. We exploit a county's river gradient and elevation to address the endogeneity of dam placement. We find that an additional dam reduces a local county's cropland net primary production (NPP) and nighttime light (NTL)-based GDP by 13.7% and 2.9%, respectively. We also find that an additional dam increases a downstream county's NPP by 0.5% and has a positive yet insignificant impact on a downstream county's NTL-based GDP. Dynamic analysis shows that the positive impact of dams on downstream counties' cropland productivity and economic development takes around ten years to realize. Co-PI He also has a working paper now under review that examines the relationship between political tensions and food import refusals: We examine the impact of political conflicts on China's food import refusals using monthly data from 2010 to 2022. The analysis reveals that political conflicts significantly contribute to increased food import rejections. Specifically, a one-standard-deviation rise in political tensions results in a 0.02% increase in the number of import rejections. These findings highlight the importance of considering the risks of shipment rejections linked to political conflicts for exporters targeting the Chinese market. Objective 2: Develop ML models, including supervised and unsupervised models, to fit and forecast China's overall agricultural production, consumption, and imports of major commodities over time Co-PI Xiong has a publication on Chinese crop yield forecasts: Crop yield forecasting is crucial for global food security. In this paper, we go beyond traditional point forecasting to examine the probability density forecasting of corn yield using a quantile-based machine learning approach. Leveraging 36 years of county-level panel data that cover 1,260 counties in China between 1980 and 2019, we develop a quantile regression forest model, which is an improvement in random forest combined with quantile regression for probability density forecasting of corn yield. Our results show that all quantile-based models produce good point forecasts, prediction intervals, and probability density curves; in general, we find that quantile regression forest with LASSO is best. Co-PIs Xiong, Hayes and PI Zhang have a working paper that is available for review: U.S. hog and pig inventory data are one of six principle economic indicators of the U.S. agricultural economy published by the National Agriculture Statistics Service. This data is published on a quarterly basis. This study proposes a dynamic factor model (DFM) to nowcast inventory values published in the quarterly Hogs and Pigs reports from 1993 to 2024 using more frequent production data from the U.S. Department of Agriculture and futures price data from the Chicago Mercantile Exchange. Our results show that the nowcasting model yields accurate predictions in the months and weeks ahead of the release of the next Hogs and Pigs report. Co-PI Hu advised a MS thesis on superstar firm prediction using ensemble machine learning and genetic algorithms, this also became the thesis by Omid Rostami: This study presents a comprehensive analysis of firm-level performance within five distinct industries, utilizing data from the Chinese Industrial Enterprises Database covering the years 2002-2007. We designed an ensemble machine learning algorithm with Random Forest, XGBoost, AdaBoost, and least absolute shrinkage and selection operator (LASSO) as the base learners coupled with a Genetic Algorithm (GA) for the optimal aggregation. Our findings reveal that "Last year's market share" consistently emerges as a significant predictor across all industries, underscoring the impact of historical performance on future market trajectory. Objective 3: Disseminate better data on China's key agricultural commodities and validated ML models via an open-source platform that readily allows collaboration, contribution, and utilization by other researchers We have started evaluating the different open-source platforms, including Mapbox, Tableau, ArcGIS StoryMaps, Wix, Github, and others that potentially be used to host our findings.

Publications

• Type: Journal Articles Status: Published Year Published: 2023 Citation: Xiong, Tao, Wendong Zhang, and Fangxiao Zhao. 2023. "When China Strikes: Quantifying Australian Companies' Stock Price Responses to China's Trade Restrictions", Australian Journal of Agricultural and Resource Economics, https://doi.org/10.1111/1467-8489.12532
• Type: Journal Articles Status: Published Year Published: 2023 Citation: He, X. (2023). Dams, cropland productivity, and economic development in China. China Economic Review, 81, 102046.
• Type: Journal Articles Status: Under Review Year Published: 2024 Citation: He, X; Jingxi, Wang (2024). Political tensions and food import refusals.
• Type: Journal Articles Status: Under Review Year Published: 2024 Citation: Tao Xiong, Zhenfeng Ma, Lee Schulz, Siyu Bian, Dermot Hayes, Wendong Zhang. 2024. Nowcasting US Hogs and Pigs Inventory.
• Type: Journal Articles Status: Published Year Published: 2023 Citation: Li, Minghao, and Xi He, Wendong Zhang, Lulu Rodriguez, James M Gbeda, Shuyang Qu. 2023. Farmers Reactions to the US-China Trade War: Perceptions Versus Behaviors. Journal of AAEA.
• Type: Journal Articles Status: Published Year Published: 2024 Citation: Mykel R. Taylor, Wendong Zhang, and Festus Attah. 2023. "Foreign Interests in U.S. Agricultural Lands: The Missing Conversations about Leasing." Choices
• Type: Journal Articles Status: Published Year Published: 2024 Citation: Wang, Fangyao, Wendong Zhang, and Mykel Taylor. 2024. Mapping and Contexualizing Foreign Ownership and Leasing of US Farmland. Journal of the ASFMRA. https://wendongzhang.weebly.com/uploads/1/4/2/2/142249534/wang_2024_jasfmra_afida.pdf
• Type: Conference Papers and Presentations Status: Published Year Published: 2024 Citation: Omid Rostami, Mohammad Fili, Jonas Swope, Guiping Hu, Wendong Zhang, Wei Zhang. Industry Superstars: Unmasking Key Features that Drive Firm-Level Performance in Chinese Markets using Ensemble Learning with Genetic Algorithm. IISE Annual Conference, 2024.
• Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: Foreign Interest in U.S. Agricultural Land. Invited Seminar for ASFMRA (American Society of Farm Managers and Rural Appraisers) National Conference, Nashville, TN, November 16, 2023 (Joint with Mykel Taylor)
• Type: Conference Papers and Presentations Status: Published Year Published: 2024 Citation: Wendong Zhang. China's Agri-Food Trends. Invited Presentation for Iowa Corn Growers Association. Via zoom, March 5, 2024.